Spatio-temporal Features for Generalized Detection of Deepfake Videos

Authors: Ipek Ganiyusufoglu, L. Minh Ngô, Nedko Savov, Sezer Karaoglu, Theo Gevers

Published: 2020-10-22 16:28:50+00:00

Comment: Submitted to Computer Vision and Image Understanding (CVIU)

AI Summary

This paper proposes using spatio-temporal features, modeled by 3D Convolutional Neural Networks (CNNs), to enhance the generalization capabilities for detecting new types of deepfake videos. The authors empirically demonstrate that existing image-level and sequence classifiers generalize poorly to unseen manipulation techniques. Their approach captures local spatio-temporal relations and inconsistencies in deepfake videos, outperforming current methods in generalization across various deepfake types and datasets.

Abstract

For deepfake detection, video-level detectors have not been explored as extensively as image-level detectors, which do not exploit temporal data. In this paper, we empirically show that existing approaches on image and sequence classifiers generalize poorly to new manipulation techniques. To this end, we propose spatio-temporal features, modeled by 3D CNNs, to extend the generalization capabilities to detect new sorts of deepfake videos. We show that spatial features learn distinct deepfake-method-specific attributes, while spatio-temporal features capture shared attributes between deepfake methods. We provide an in-depth analysis of how the sequential and spatio-temporal video encoders are utilizing temporal information using DFDC dataset arXiv:2006.07397. Thus, we unravel that our approach captures local spatio-temporal relations and inconsistencies in the deepfake videos while existing sequence encoders are indifferent to it. Through large scale experiments conducted on the FaceForensics++ arXiv:1901.08971 and Deeper Forensics arXiv:2001.03024 datasets, we show that our approach outperforms existing methods in terms of generalization capabilities.

Key findings

Spatio-temporal features obtained from 3D CNNs achieve superior generalization to different deepfake methods compared to models relying on spatial or sequential features. This is because spatio-temporal features capture similarities between deepfake methods, whereas spatial features learn method-specific attributes. 3D CNNs effectively model spatio-temporal correlation and detect local disruptions, unlike sequence encoders which are largely indifferent to temporal order and inconsistencies.

Approach

The authors propose detecting deepfake videos by utilizing spatio-temporal features modeled by 3D Convolutional Neural Networks (CNNs). This method captures local spatio-temporal relations and inconsistencies by convolving 3D kernels over consecutive video frames, learning a joint representation of both spatial and temporal dimensions simultaneously.

Datasets

FaceForensics++, Deepfake Detection Challenge (DFDC), Deeper Forensics

Model(s)

R3D-18, I3D-RGB, XceptionNet, EfficientNet-B3, EfficientNet-B3 + LSTM, EfficientNet-B3 + bi-directional GRU

Author countries

The Netherlands

← Previous